Supplementary description for NKS application. Failure management in nuclear power plants. Research topic and justification

Transcription

1 Supplementary description for NKS application Failure management in nuclear power plants Research topic and justification The aim is early detection of faults in nuclear power plants, and to support operators and experts in their decision making. Neural methods, such as Self-Organizing Map (SOM), are used in combination with more traditional methods. We plan to develop various visualizations to support the decision making in the control room, and to ennoble the information content of the operator. We will use existing data-analysis methodologies, and we will also develop assisting methodologies for this purpose. This is an important research area, which also power companies are interested in. One need is to develop new contents to big monitoring screens in the modern control room. These are needed both in new nuclear power plants, and also in old ones, which are going through a modernization project. Our visualizations can be used both in big screens as well as in ordinary workstation screens. We are developing new presentation techniques and methods for ennobling information for the operator. Our methodologies can also be used for instance in training simulator behavior assessment. The data-analysis methodologies are utilized here. For example, if a change is made to the process, after that it is possible to analyze how well the simulator adapts to the changed circumstances. The aim is also to initialize co-operation with Halden Reactor Project (HRP) and with a psychology group in Technical Research Centre (VTT) with similar or related research area. We also plan to continue the ongoing co-operation with the Finnish power companies. Background We have been developing a decision support prototype for several years. This control room tool is based on knowledge-based methods and neural methods. It is able to give advice for the operator or expert in textual form, and it also produces many kind of visual decision support for them. We have also carried out data analysis with nuclear power plant data and training simulator in co-operation with Finnish industry. Last three years this work has mainly been done in an industrial project NoTeS/NoTeS2. Many research partners and companies have been involved. The project is a part of the Finnish TEKES technology program MASI (modeling and simulation). This project will end in the end of this year. In our subproject the industrial partner has been TVO Olkiluoto. An initial version of our decision support prototype will be installed in TVO Olkiluoto for test use. Goals and results The main goals are to help the operators and experts in early fault detection and provide them with methodological and visual aid in their decision making by using data-analysis methodologies in combination with more conventional methods. In addition we try to help the end users in some of their specific problems in the plant by using our

2 methodologies and techniques. Our goal is also to begin some international co-operation on this area. The results will be in the form of program code and analysis reports, which will also introduce our tools and methods that we are using in solving various problems, and therefore reflect also our point of view in research. On the way we will also educate new young scientists on this area. We will also write some international publications of the research results of the project, and on the long run at least one PhD thesis will be written. Utilization of the results and the relevance to the program The results of this project can be utilized by the end users, mainly power companies. Also product suppliers will get new ideas what kind of tools to develop to the control room of a nuclear power plant. All research partners connected to the project can utilize the results, as well as all the Nordic authorities. The proposed project has many values supporting the NKS criteria. The project will add Nordic value by increasing the co-operation with different Nordic Research Institutes, and it makes it also possible in future to co-operate with nuclear industries all over the Scandinavia. The research will be done according to high technical and scientific standards, because good experts are used both in the research side and in industrial side. As results also international publications will be written, which are available for all research communities all over the world. Also a master s thesis and a PhD thesis are in our plans. The proposed project is relevant to NKS end-users, because the research topic is both important and according to current interest at the moment. The methodologies and tools that we develop will add value to the control room equipment, and will also help in assessing the current and future equipment. At the moment we have in our current project one Master s student and one PhD student. We have also had a couple of younger students in the project on the way. In addition we have had one senior supervisor and one or two younger instructors in the previous projects. In the proposed project we are planning to go on the work with similar team. Therefore participation of young students is a very strong point in our project proposal. We are just ending a large national research program with many research institutes and many industrial partners. So we have very many ready national contacts that can be utilized also in the new project. We have had some international contacts as well, for instance with the OECD Halden Reactor Project (HRP). We are planning to deepen both our national and international contacts in the new project, and initialize co-operation, which could make it possible to co-operate more closely in the future, especially with VTT and HRP. Work plan The work plan consists of two separate tasks:

3 1) Detecting pre-stage of process fault by data-driven methods 2) Development of a decision support tool in NPP control room The two different tasks have some interconnection. The first task is part of a PhD study. The second task is continuation for a master s thesis that will be finished next winter. The first task is the main focus in the project funded partly by NKS. In co-operation with VTT and HRP we are going to assess the usability of the visualizations developed in the project, and we will compare them with other existing visualizations that are either already in use in existing plants or at development state in the HRP research laboratory. We are also going to observe what additional value the new visualization methods are able to bring in informing operators. Detecting pre-stage of process fault by data-driven methods Data-driven fault detection process monitoring methods are widely used, because of their simplicity and clarity. Ignorance of correlations between different process signals is a deficiency in these methods. The need to handle spatial correlations in high-dimensional processes has led to the development of multivariate statistical techniques. We plan to use these methods in adequate combination. Principal Component Analysis (PCA) and Partial Least Squares (PLS) are most commonly used dimensionality reduction methods. PCA is an optimal dimensionality reduction technique, which is used to reduce the complexity of the representation, but also for selection of model features. PLS is maximizing covariance between dependent variables and interpretative variables. The nonlinear behavior in dynamic processes gives even more challenges. In our project these methods are planned to be used. Our objectives in this research are to develop methods in fault detection, and also to provide operator with better tools in getting this information. Safety is the most important criterion, but also optimizing costs and power production are issues that will be covered. Adaptive models help in analyzing the process step by step. The delays between different variables in real failures are challenging in data analysis. The segmentation method need to be selected differently in different data structures depending on the time span. Variable dependencies in industrial processes are important to take into account as well. One further goal is to combine different methods used in different phases of the project, and make useful combinations out of them. The control room modernization projects as well as the new control rooms of completely new nuclear power plants have rose up new visualization challenges. For instance, how to use wide monitoring screens efficiently to help operators in their decision making. When fault or its pre-stage is detected, and current process state is diagnosed, operator should also be informed efficiently. Fault diagnosis is determining the type, location, magnitude and time of the fault. Early detection gives an invaluable warning on the problem. Development of a decision support tool in NPP control room

4 Development of a decision support tool in nuclear power plant control room is the smaller subtask in this project. It is continuation of similar work carried out in our earlier project introduced earlier in this application. The tool for operators and experts is now in a phase of more industrial test use. It is also a possible platform for further experimental testing of various methods in this project. In this use the control room tool can also be further developed itself. Expense and financing plan The amount of the total project for one year is DKK. Half of the money DKK will be applied from NKS, and the other half will be financed by Helsinki University of Technology (TKK) and the industrial partners. Most of the project budget will be used in the salaries of two PhD students. About 20% of the budget is reserved for materials, traveling and other expenses. Project organization The project organization is the following: Miki Sirola, D.Sc. (Tech.), Docent Jukka Parviainen, M.Sc. (Tech.) Jaakko Talonen, M.Sc. (Tech.) Golan Lampi, M.Sc. (Tech.) student Dr. Miki Sirola is responsible for the project management and supervision in the research group. Mr. Jukka Parviainen is another supervisor in the group. Mr. Jaakko Talonen and Mr. Golan Lampi are mainly carrying out the research work. Possibly a summer student will join the research team next summer. The whole research group is working in the Department of Information and Computer Science in the Helsinki University of Technology. References Sirola, Miki; Lampi, Golan; Parviainen Jukka. SOM based decision support in failure management. International Scientific Journal of Computing, 2005, Vol. 3, Issue 4, pp Sirola, Miki. New methodologies and visualization techniques for NPP control room concept. Workshop on Human System Interfaces - Design and Evaluation (HSI). Halden, Norway, 4-5 May OECD Halden Reactor Project. 5 p. Hakala, Risto; Similä, Timo; Sirola, Miki; Parviainen, Jukka. Process state and progress visualization using self-organizing map. International Conference on Intelligent Data Engineering and Automated Learning (IDEAL'2006). Burgos, Spain, September p. Ritala, Risto; Alhoniemi, Esa; Kauranne, Tuomo; Konkarikoski, Kimmo; Lendasse, Amaury; Sirola, Miki. Nonlinear temporal and spatial forecasting: modeling and uncertainty analysis (NoTeS) - MASIT20. MASI Technology Programme Yearbook Tekes Technology Review 207/ p.

5 Sirola, Miki; Lampi, Golan; Parviainen, Jukka. Failure detection and separation in SOM based decision support. Workshop on Self-Organizing Maps (WSOM 2007). Bielefeld, Germany, 3-6 September p. Alhonnoro, Tuomas; Sirola, Miki. Feature selection on process fault detection and visualization. European Symposium on Artificial Neural Networks (ESANN 2008). Bruges, Belgium, April p. Sirola, Miki; Parviainen, Jukka; Talonen, Jaakko; Lampi, Golan; Alhonnoro, Tuomas; Hakala, Risto. Early fault detection with SOM based methods and visualizations - new contents for wide monitoring screens. EHPG-Meeting of OECD Halden Reactor Project. Loen, Norway, May p. Talonen, Jaakko; Sirola, Miki; Parviainen, Jukka. Leakage detection by adaptive process modeling. International Conference on Data Mining (DMIN 08). Las Vegas, USA, July. 4 p. Ritala, Risto; Alhoniemi, Esa; Kauranne, Tuomo; Konkarikoski, Kimmo; Lendasse, Amaury; Sirola, Miki. Nonlinear temporal and spatial forecasting: modeling and uncertainty analysis (NoTeS) - MASIT20. MASI Programme Yearbook Tekes Review 228/ p.